View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Elsevier - Publisher Connector Discrete Mathematics 306 (2006) 867–875 www.elsevier.com/locate/disc Normal hypergraphs and the perfect graph conjecture L. Lovász Chair of Geometry, Eotvös Loránd University, Budapest, Hungary Abstract A hypergraph is called normal if the chromatic index of any partial hypergraph H of it coincides with the maximum valency in H . It is proved that a hypergraph is normal iff the maximum number of disjoint hyperedges coincides with the minimum number of vertices representing the hyperedges in each partial hypergraph of it. This theorem implies the following conjecture of Berge: The complement of a perfect graph is perfect. A new proof is given for a related theorem of Berge and Las Vergnas. Finally, the results are applied on a problem of integer valued linear programming, slightly sharpening some results of Fulkerson. © 1972 Published by Elsevier B.V. m0. Introduction Let G be a finite graph and let (G) and (G) denote its chromatic number and the maximum number of vertices forming a clique in G, respectively. Obviously, (1) (G) ≥ (G). There are several classes of graphs such that (2) (G) = (G), e.g., bipartite graphs, their line graphs and complements, interval graphs, transitively orientable graphs, etc. Obviously, relation (2) does not say too much about the structure of G; e.g. adding a sufficiently large clique to an arbitrary graph, the arising graph satisfies (1). Berge [1,2] has introduced the following concept: a graph is perfect (-perfect) if the equality holds in (2) for every induced subgraph of it. The mentioned special classes of graphs have this property, since every induced subgraph of them belongs to the same class. He formulated two conjectures in connection with this notion: Conjecture 1. A graph is perfect if and only if neither itself nor its complement contains an odd circuit without diagonals. DOI of original article: 10.1016/0012-365X(72)90006-4 The original article was published in Discrete Mathematics 2 (1972) 253–267 0012-365X/$ - see front matter © 1972 Published by Elsevier B.V. doi:10.1016/j.disc.2006.03.007 868 L. Lovász / Discrete Mathematics 306 (2006) 867–875 Conjecture 2. Let (G) denote the stability number of G, let (G) denote the minimum number of cliques which partition the set of all the vertices. A graph G is perfect if and only if (G) = (G) for any induced subgraph of G. This conjecture is an attempt to explain some similarities between the properties of the chromatic number and the stability number; his next conjecture is proved in the present paper, formulated as follows. Perfect graph theorem. The complement of a perfect graph is perfect as well. Obviously, the second conjecture of Berge would follow from the first one. However, due to its simpler form, it has more interesting applications and has been more investigated. Partial results are due to Berge [3], Berge and Las Vergnas [4], Sachs and Olaru [6]. Fulkerson [5] reduced the problem to the following conjecture, using the theory of anti-blocking polyhedra: Duplicating an arbitrary vertex of a perfect graph and joining the obtained two vertices by an edge, the arising graph is perfect. In m1 we prove a theorem which contains this conjecture. Berge has observed that the perfect graph conjecture has an equivalent in hypergraph theory, interesting for its own sake too. The correspondence between graphs and hypergraphs is simple and enables us to translate proofs formulated in terms of graphs into proofs with hypergraphs and conversely. In m2 we deduce the hypergraph version of the perfect graph theorem from the above-mentioned conjecture of Fulkerson; the proof is short and does not use the theory of anti-blocking polyhedra. It could be formulated in terms of graphs as well; however, the hypergraph-version shows the idea more clearly. It should be pointed out that thus the proof consists of two steps and the more difficult second step was done first by Fulkerson. In m3, we give a new proof of a related theorem of Berge. Finally, in m4 we give some formulations of the results in terms of linear programming. Most of them have been observed to be equivalent with the perfect graph theorem already by Fulkerson. m1 Let G, H be two vertex-disjoint graphs and let x be a vertex of G. By substituting H for x we mean deleting x and joining every vertex of H to those vertices of G which have been adjacent with x. Theorem 1. Substituting perfect graphs for some vertices of a perfect graph the obtained graph is also perfect. Proof. We may assume that only one perfect graph H is substituted for a vertex x of a perfect graph G. Let G be the resulting graph. It is enough to show that (3) (G ) = (G ), since for the induced subgraphs of G, which arise by the same construction from perfect graphs, this follows similarly. We use induction on k = (G).Fork = 1 the statement is obvious. Assume k>1. It is enough to find a stable set T of G meeting all k-element cliques, since then coloring these vertices by the same color and the remaining vertices by k − 1 other colors (which can be done by the induction hypothesis), we obtain a k-coloring of G. Put m = (G), n = (H ), and let p denote the maximum cardinality of a clique of G containing x. Then, obviously, k = max{m, n + p − 1}. Consider an m-coloring of G and let K be the set of vertices having the same color as x. Let, further, L be a set of independent vertices of H meeting every n-element clique of H. Then T = L ∪ (K\{x}) is a stable set in G. Moreover, T intersects every k-element clique of G. Really, if C is a k-element clique of G and it meets H then, obviously, it contains an n-element clique of H and thus a vertex of L. On the other hand, if C does not meet H, then C must be an m-element clique of G, and thus C contains a vertex of K\{x}. As it has been mentioned in the introduction, in view of Fulkerson’s results, the perfect graph theorem already follows from Theorem 1. However, to make the paper self-contained, we give a proof of the perfect graph conjecture (which seems to be different from that of Fulkerson). L. Lovász / Discrete Mathematics 306 (2006) 867–875 869 m2 A hypergraph is a non-empty finite collection of non-empty finite sets called edges. The elements of edges are the vertices. Multiple edges are allowed, i.e. more (distinguished) edges may have the same set of vertices. The number of edges with the same vertices is called the multiplicity of them. The number of edges containing a given vertex is the degree of it. The maximum degree of vertices of a hypergraph H will be denoted by (H ). A partial hypergraph of H is a hypergraph consisting of certain edges of H. The subhypergraph spanned by a set X of vertices means the hypergraph H|X ={E ∩ X|E ∈ H,E ∩ X =∅}. A partial subhypergraph is a subhypergraph of a partial hypergraph (or, equivalently, a partial hypergraph of a subhy- pergraph). The chromatic number (H ) of a hypergraph H is the least number of colors sufficient to color the vertices (so that every edge with more than one vertices has at least two vertices with different colors). The chromatic index (H ) of H is the least number of colors by which the edges can be colored so that edges with the same color are disjoint. Obviously, (4) (H ) ≥ (H ). Let a hypergraph be called normal if the equality holds in (4) for every partial hypergraph of it. A set T of vertices of H is called a transversal if it meets every edge of H ; (H ) is the minimum cardinality of transversals. Denoting by (H ) the maximum number of pairwise disjoint edges of H, we obviously have (5) (H ) ≤ (H ). Let a hypergraph be called -normal if the equality holds in (5) for every partial hypergraph of it. A hypergraph is said to have the Helly property if any collection of edges whose intersection is empty contains two disjoint edges. It is easily seen that normal and -normal hypergraphs have the Helly property. Given a hypergraph H, we can consider its edge-graph G(H ) defined as follows: the vertices of G(H ) are the edges of H and two edges of H are joined iff they intersect. On the other hand, for a given graph G we can construct a hypergraph H (G) by considering the maximal cliques of G (in the set-theoretical sense) as vertices of H and, for any vertex x of G, the set of maximal cliques containing x, as an edge of H (G). It is easily shown that if G has no multiple edges (which can be assumed throughout this paper) then (6) G(H (G)) G. Furthermore, H (G) always has the Helly property. It is easily seen that (7) (G(H )) = (H ), (G(H )) = (H ) (G(H ) is the complement of G(H )). Moreover, if H has the Helly property then (8) (G(H )) = (H ), (G(H )) = (H ). Hence by (6), (9) (G) = (H (G)), (G) = (H (G)), (G)¯ = (H (G)), (G)¯ = (H (G)), for any graph G. Equalities (7), (8) and (9) imply Theorem 2. H is normal iff G(H ) is perfect; G is perfect iff H (G) is normal.